Presentation is loading. Please wait.

Presentation is loading. Please wait.

Kivik 2013Kilgarriff: Web corpora1 Web Corpora Adam Kilgarriff.

Similar presentations


Presentation on theme: "Kivik 2013Kilgarriff: Web corpora1 Web Corpora Adam Kilgarriff."— Presentation transcript:

1 Kivik 2013Kilgarriff: Web corpora1 Web Corpora Adam Kilgarriff

2 Kivik 2013Kilgarriff: Web corpora2 You can’t help noticing Replaceable or replacable? –http://googlefight.comhttp://googlefight.com

3 Kivik 2013Kilgarriff: Web corpora3 Very very large –2006 estimates for duplicate free, linguistic, Google- indexed web German: 44 billion words Italian: 25 billion words English: 1,000 billion -10,000 billion words Most languages Most language types Up-to-date Free Instant access

4 Kivik 2013Kilgarriff: Web corpora4 Overview Is the web a corpus? Representativeness What is out there? –Web1T Googleology Web corpus types –Targeted sites: Oxford English Corpus –General: WaC family –WebBootCaT

5 Kivik 2013Kilgarriff: Web corpora5 Is the web a corpus? Sinclair –in “Developing linguistic corpora, a guide to good practice. Corpus and Text – Basic Principles” “…not a corpus because dimensions unknown, constantly changing not designed from a linguistic perpective But –We can find out dimensions –Many corpora are not designed “as much chatroom dialogue as I can get” Def: a corpus is a collection of texts –when viewed as an object of language research

6 Kivik 2013Kilgarriff: Web corpora6 Is the web a corpus? Yes

7 Kivik 2013Kilgarriff: Web corpora7 but it’s not representative

8 Kivik 2013Kilgarriff: Web corpora8 Theory A random sample of a population is representative of it. Observations on sample support inferences about population (within confidence bounds)‏

9 Kivik 2013Kilgarriff: Web corpora9 Theory A random sample of a population is … What is the population? –production and reception –speech and text –copying

10 Kivik 2013Kilgarriff: Web corpora10 Theory Population not defined Representative sample not possible

11 Kivik 2013Kilgarriff: Web corpora11 sublanguage Language = core + sublanguages Options for corpus construction –none –some –all None –impoverished view of language Some: BNC –cake recipes and gastro-uterine disease –not car repair manuals or astronomy or … All: until recently, not viable

12 Kivik 2013Kilgarriff: Web corpora12 Representativeness The web is not representative but nor is anything else Text type variation –under-researched, lacking in theory Atkins Clear Ostler 1993 on design brief for BNC; Biber 1988, Kilgarriff 2001 Text type is an issue across NLP –Web: issue is acute because, as against BNC or WSJ, we simply don’t know what is there

13 Kivik 2013Kilgarriff: Web corpora13 What is out there? What text types are there on the web? –some are new: chatroom –proportions is it overwhelmed by porn? How much? Hard question

14 Kivik 2013Kilgarriff: Web corpora14 The web –a social, cultural, political phenomenon –new, little understood –a legitimate object of science –mostly language we are well placed –a lot of people will be interested Let’s –study the web –source of language data –apply our tools for web use (dictionaries, MT)‏ –use the web as infrastructure

15 Kivik 2013Kilgarriff: Web corpora15 Using Search Engines No setup costs Start querying today Methods Hit counts ‘snippets’ –Metasearch engines, WebCorp Find pages and download

16 Kivik 2013Kilgarriff: Web corpora16 Googleology Google hit counts for language modelling –Example: (Keller & Lapata 2003) –36 queries to estimate freq(fulfil, obligation) to each of Google and Altavista Very interesting work Great interest in query syntax

17 Kivik 2013Kilgarriff: Web corpora17 The Trouble with Google not enough instances –max 1000 not enough queries –max 1000 per day with API not enough context –10-word snippet around search term sort order –search term in titles and headings untrustworthy hit counts limited search options linguistically dumb, eg not lemmatised aime/aimer/aimes/aimons/aimez/aiment …

18 Kivik 2013Kilgarriff: Web corpora18 Appeal –Zero-cost entry, just start googling Reality –High-quality work: high-cost methodology

19 Kivik 2013Kilgarriff: Web corpora19 Also: No replicability Methods, stats not published At mercy of commercial corporation Googleology is bad science

20 Kivik 2013Kilgarriff: Web corpora20 Better: web-sourced corpora Gather pages –Google hits –Select and gather whole sites –General crawl Filter De-duplicate Linguistic processing Load into corpus tool

21 Kivik 2013Kilgarriff: Web corpora21 Oxford English Corpus Whole domains chosen and harvested –control over text type 2.3 billion words

22 Kivik 2013Kilgarriff: Web corpora22 Oxford English Corpus

23 Kivik 2013Kilgarriff: Web corpora23 WaC family 1.5 B words each Baroni and colleagues Seeds: –mid-frequency words from ‘core vocab’ lists and corpora Google on seed words, then crawl

24 TenTen Family Processing chain –Spiderling, a lingustic crawler A billion words a day –jusText for“cleaning”: removing non-text –Onion – remove duplicates (paragraph level) All major world languages 2-20 billion words Lexical Computing All available in Sketch Engine Kivik 2013Kilgarriff: Web corpora24

25 Kivik 2013Kilgarriff: Web corpora25 Small, specialised corpora Terminologists Translators needing target-language domain-specific vocab Specialist dictionaries –Don’t exist –Expensive/inaccessible –Out of date

26 Kivik 2013Kilgarriff: Web corpora26 BootCat ( Bootstrapping Corpora and Terms) Put in seed terms Google/Yahoo search Retrieve Google/Yahoo hits –Remove duplicates, boilerplate Small instant corpora Baroni and Bernardini, LREC 2004 Web version –WebBootCaT –At Sketch Engine site

27 But did I make a good corpus? Kivik 2013Kilgarriff: Web corpora27

28 Bad Science Ben Goldacre Kivik 2013Kilgarriff: Web corpora28

29 Bad Science Ben Goldacre Biases in samples –A quarter of the people who tested positive had just been on holiday in Mexico –But the research team didn’t notice Kivik 2013Kilgarriff: Web corpora29

30 Bad linguistics Our corpus study shows X –But what was in the corpus? Kivik 2013Kilgarriff: Web corpora30

31 Bad linguistics Our corpus study shows X –But what was in the corpus? –Moral: Get to know your corpus Kivik 2013Kilgarriff: Web corpora31

32 How? Read it? Too big to read Not designed to be read Kivik 2013Kilgarriff: Web corpora32

33 How? Compare it with other(s) Keyword lists Kivik 2013Kilgarriff: Web corpora33

34 UKWaC vs. enTenTen12 Kivik 2013Kilgarriff: Web corpora34

35 enTenTen vs. UKWaC accord actually amendment among bad because behavior believe bill blog ca center citizen color defense determine do dollar earth effort election even evil fact faculty favor favorite federal foreign forth guess guy he her him himself his honor human kid kill kind know labor law let liberal like man maybe me military movie my nation never nor not nothing official oh organization percent political post president pretty professor program realize recognize say shall she sin soul speak state suppose tell terrorist that thing think thou thy toward true truth unto upon violation vote voter war what while why woman yes accommodation achieve advice aim area assessment available band behaviour building centre charity click client club colour consultation contact council delivery detail develop development disabled email enable enquiry ensure event excellent facility favourite full further garden guidance guide holiday improve information insurance join link local main manage management match mm nd offer opportunity organisation organise page partnership please pm poker pp programme project pub pupil quality range rd realise recognise road route scheme sector service shop site skill specialist st staff stage suitable telephone th top tour training transport uk undertake venue village visit visitor website welcome whilst wide workshop www Kivik 2013Kilgarriff: Web corpora35

36 enTenTen vs. UKWaC accord actually amendment among bad because behavior believe bill blog ca center citizen color defense determine do dollar earth effort election even evil fact faculty favor favorite federal foreign forth guess guy he her him himself his honor human kid kill kind know labor law let liberal like man maybe me military movie my nation never nor not nothing official oh organization percent political post president pretty professor program realize recognize say shall she sin soul speak state suppose tell terrorist that thing think thou thy toward true truth unto upon violation vote voter war what while why woman yes accommodation achieve advice aim area assessment available band behaviour building centre charity click client club colour consultation contact council delivery detail develop development disabled email enable enquiry ensure event excellent facility favourite full further garden guidance guide holiday improve information insurance join link local main manage management match mm nd offer opportunity organisation organise page partnership please pm poker pp programme project pub pupil quality range rd realise recognise road route scheme sector service shop site skill specialist st staff stage suitable telephone th top tour training transport uk undertake venue village visit visitor website welcome whilst wide workshop www Kivik 2013Kilgarriff: Web corpora36

37 enTenTen vs. UKWaC accord actually amendment among bad because behavior believe bill blog ca center citizen color defense determine do dollar earth effort election even evil fact faculty favor favorite federal foreign forth guess guy he her him himself his honor human kid kill kind know labor law let liberal like man maybe me military movie my nation never nor not nothing official oh organization percent political post president pretty professor program realize recognize say shall she sin soul speak state suppose tell terrorist that thing think thou thy toward true truth unto upon violation vote voter war what while why woman yes accommodation achieve advice aim area assessment available band behaviour building centre charity click client club colour consultation contact council delivery detail develop development disabled email enable enquiry ensure event excellent facility favourite full further garden guidance guide holiday improve information insurance join link local main manage management match mm nd offer opportunity organisation organise page partnership please pm poker pp programme project pub pupil quality range rd realise recognise road route scheme sector service shop site skill specialist st staff stage suitable telephone th top tour training transport uk undertake venue village visit visitor website welcome whilst wide workshop www Kivik 2013Kilgarriff: Web corpora37

38 enTenTen vs. UKWaC accord actually amendment among bad because behavior believe bill blog ca center citizen color defense determine do dollar earth effort election even evil fact faculty favor favorite federal foreign forth guess guy he her him himself his honor human kid kill kind know labor law let liberal like man maybe me military movie my nation never nor not nothing official oh organization percent political post president pretty professor program realize recognize say shall she sin soul speak state suppose tell terrorist that thing think thou thy toward true truth unto upon violation vote voter war what while why woman yes accommodation achieve advice aim area assessment available band behaviour building centre charity click client club colour consultation contact council delivery detail develop development disabled email enable enquiry ensure event excellent facility favourite full further garden guidance guide holiday improve information insurance join link local main manage management match mm nd offer opportunity organisation organise page partnership please pm poker pp programme project pub pupil quality range rd realise recognise road route scheme sector service shop site skill specialist st staff stage suitable telephone th top tour training transport uk undertake venue village visit visitor website welcome whilst wide workshop www Kivik 2013Kilgarriff: Web corpora38

39 enTenTen vs. UKWaC Core verbs –be determine do guess know let say shall suppose tell think Pronouns –he her him his me my she Biber: more informal Kivik 2013Kilgarriff: Web corpora39

40 Judgements Not all or nothing –Both have (lots of) AmE and BrE –Observing patterns Not right or wrong Where does ‘believe’ belong? –Bible or core verbs? –No right answer, could be both The better you know the data, the better you understand why words are there Kivik 2013Kilgarriff: Web corpora40

41 The maths “this word is twice as common here as there” Simplest approach –Normalise frequencies Per thousand, or per million –Take ratio For examples –Assume two 1m-word corpora Normalisation not needed –Fc=focus corpus –Rc= reference corpus Kivik 2013Kilgarriff: Web corpora41

42 Kivik 2013Kilgarriff: Web corpora42 Problem 1: You can’t divide by zero Standard solution: add one Problem solved fc rcratio buggle100? stort1000? nammikin10000? fc rcratio buggle111 stort1011 nammikin10011

43 Kivik 2013Kilgarriff: Web corpora43 Problem 2: High ratios more common, less interesting for rarer words fc rc ratiointeresting? spug101 no grod100010010yes ratio is not enough: frequency matters too Also some researchers: grammar, grammar words some researchers: lexis, content words No right answer Slider?

44 Kivik 2013Kilgarriff: Web corpora44 Solution Don’t just add 1, add n: n=1 n=100 word fc rc fc+n rc+nRatioRank obscurish10011111.001 middling2001002011011.992 common120001000012001100011.203 word fc rc fc+n rc+nRatioRank obscurish1001101001.103 middling2001003002001.501 common120001000012100101001.202

45 Kivik 2013Kilgarriff: Web corpora45 n=1000 word fc rc fc+n rc+nRatioRank obscurish100101010001.013 middling200100120011001.092 common120001000013000110001.181

46 Kivik 2013Kilgarriff: Web corpora46 Summary word fc rc n=1 n=100n=1000 obscurish1001st2nd3rd middling2001002nd1st2nd common12000100003rd 1st

47 Kivik 2013Kilgarriff: Web corpora47 But what about Mutual information Log-likelihood Chi-square Fisher’s test … Don’t they use cleverer maths?

48 Kivik 2013Kilgarriff: Web corpora48 Yes but Clever maths is for hypothesis testing –Can you defeat null hypothesis? Language is not random, so … you always can Null hypothesis never true Hypothesis-testing not informative Clever maths irrelevant –Kilgarriff 2006, CLLT

49 Kivik 2013Kilgarriff: Web corpora49 Varying the parameter BAWE –British Academic Written English Nesi and Thompson 2008 –Student essays Arts/Humanities, Social Sciences, Life Sciences, Physical Sciences – fc: ArtsHum, rc: SocSci –With n=10 and n=1000

50 Kivik 2013Kilgarriff: Web corpora50

51 Kivik 2013Kilgarriff: Web corpora51

52 Parameters for keyword lists Lemmas –Could be word forms, word classes Simplemaths –(default: 100, for mix of lexical and grammar words) Only all-lowercase-letters –Could allow uppercase, or any at all Minimum 2/3/4 characters –Helps get words, not abbreviations etc Kivik 2013Kilgarriff: Web corpora52

53 enTenTen vs. UKWaC Obama Clinton Hillary McCain Centre Leeds Manchester Edinburgh Kivik 2013Kilgarriff: Web corpora53 With parameters: Simplemaths: 10 Uppercase and lowercase Minimum length =5 (to exclude acronyms)

54 Two interlocking questions How do two corpora differ How do two text types differ Kivik 2013Kilgarriff: Web corpora54

55 Two interlocking questions How do two corpora differ –enTenTen vs. UKWaC –Interpret as: Differences of corpus compilation procedures - and/or - Differences of proportions of text types How do two text types differ –BAWE example Arts/humanities essays vs. Social Sciences essays –Any other corpus differences Unwanted biases But we need to know about them Kivik 2013Kilgarriff: Web corpora55

56 Don’t do bad science Get to know your corpus –Compare with others Qualitatively: keyword lists (Quantitatively: distances) No excuses –The Sketch Engine does all the technical work for you The joy of research Kivik 2013Kilgarriff: Web corpora56


Download ppt "Kivik 2013Kilgarriff: Web corpora1 Web Corpora Adam Kilgarriff."

Similar presentations


Ads by Google